MagiLLM is an on-device LLM runtime plugin for Unreal Engine. It lets you run GGUF language models locally for NPC dialogue, gameplay conversations, and offline AI assistants without cloud APIs, API keys, or per-message costs.

Features:

- Local inference powered by llama.cpp

- Blueprint-callable engine subsystem

- Streaming token events with OnToken and OnDone

- Runtime persona and language switching

- Console commands for quick testing

- Bundled Gemma GGUF models for zero-setup chat

- Supports custom GGUF models

- Win64 runtime support

Use MagiLLM to prototype AI NPCs, generate dynamic dialogue, build offline in-game assistants, or add local generative text features to Unreal Engine projects.

This plugin includes source code, llama.cpp integration, model loading helpers, multilingual persona files, and bundled model license notices. The included Gemma model weights are distributed under Google's Gemma Terms of Use. Gemma is a trademark of Google LLC. This product is not endorsed by or affiliated with Google.